A Multivariate Analysis of Forest Communities in the Western Great Smoky
Mountains National Park
Ragan M. Callaway; Edward E. C. Clebsch; Peter S. White
American Midland Naturalist, Vol. 118, No. 1. (Jul., 1987), pp. 107-120.
Stable URL:
http://links.jstor.org/sici?sici=0003-0031%28198707%29118%3A1%3C107%3AAMAOFC%3E2.0.CO%3B2-1
American Midland Naturalist is currently published by The University of Notre Dame.
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/about/terms.html. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained
prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in
the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/journals/notredame.html.
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
The JSTOR Archive is a trusted digital repository providing for long-term preservation and access to leading academic
journals and scholarly literature from around the world. The Archive is supported by libraries, scholarly societies, publishers,
and foundations. It is an initiative of JSTOR, a not-for-profit organization with a mission to help the scholarly community take
advantage of advances in technology. For more information regarding JSTOR, please contact [email protected].
http://www.jstor.org
Sun Aug 19 16:00:21 2007
A Multivariate Analysis of Forest Communities in the Western Great Smoky Mountains National Park RAGAN M . CALLAWAY1 and EDWARD E. C . CLEBSCH
Graduate Program i n Ecology, University of Tennessee, Knoxville 3 7 9 9 6
and
P E T E R S. W H I T E
Uplands Field Research Laboratory, Great Smoky Mountains National Park, Gatlinbura Tennessee 3 7 7 3 8
ABSTRACT:
An integrated sequence of multivariate techniques was applied to the
vegetation of the western Great Smoky Mountains National Park. Included in the sequence were hierarchial classification, detrended correspondence analysis and multiple discriminant analysis. Based on the importance of landform in previous research,
a system of topographical quantification was developed and also used in the classification.
Elevation and the topographical index, termed protection, were the most important variables associated with the 12 classified forest types. Drainage area, another
topographic measurement, soil p H and water-holding capacity were also significant
in the vegetation analysis. Logging and farming disturbances prior to the formation
of the national park were probable causal factors for some forest types.
INTRODUCTION
T h e Great Smoky Mountains of Tennessee and North Carolina have been a focal
point of vegetation analysis since 1930 (Cain, 1930). Because the Smoky Mountains
provide elevation-related temperature and precipitation gradients, complex topographical patterns, and a rich woody-species composition, they have been used repeatedly in
studies of the relationships of plants to the environment and within communities.
Whittaker (1956) identified species associations throughout the Great Smoky
Mountains National Park (GRSM) and defined their positions on a moisture complexgradient. H e concluded that the interaction of elevation, topographic shape and slope
aspect determined the vegetation present at a given site. Golden (1974, 1981) approached the problem of vegetation distribution in the central G R S M with an integrated sequence of multivariate techniques (cluster analysis and similarity sorting, discriminant analysis, reciprocal averaging, ordination and canonical correlation) using
quantitative environmental variables. H e found the overwhelming importance of topography and elevation as described by Whittaker was also indicated by his data and analyses. Soils data collected by Golden (1981) indicated that the p H of the A horizon and
the percent clay in the B horizon were also important correlative factors.
T h e major purposes of this paper are (1) to develop a method for quantifying topographic shape in the Smoky Mountains; (2) to provide an integrated multivariate analysis, including the topographic index, of the western Great Smoky Mountains, and (3)
to examine stands disturbed by prepark logging and agriculture. Neither Whittaker
(1956) nor Golden (1981) satisfactorily quantified topographic shape, which they cited
as being important. Whittaker (1956) assigned a sequential order to land shapes and assumed that the order represented moisture characteristics. Golden (1981) measured the
position of stands on slopes and the slope aspect but did not include slope shape. We
have constructed a n index of topographic shape and included it in our vegetation analysis. Most of the study area was on three major topographic features in the western
'Present address: Department of Biological Sciences, University of California, Santa Barbara,
CA 93106.
107
GRSM (Davis, Leadbetter and Gregory ridges) which had not been included in any
previous vegetation analyses. The plots ranged from 300-1800 m in elevation. The
spruce-fir forests of Whittaker (1956) and Golden (1981) are not present in the western
GRSM. This study also extended the data base from the undisturbed stands of Whittaker (1956) and Golden (1981) to include forests that had been disturbed by logging and
farming, thus permitting a comparison between disturbed and undisturbed stands. It
should be pointed out that the chestnut blight (Endothia parmitica) has affected almost all
forest communities in the GRSM.
METHODS
Vietation distribution.-Data from 154 permanent 20 x 50 m (0.1 ha) plots previously
established by the National Park Service were used to assess vegetation distribution.
These plots had been established with the intent to represent all combinations of elevational, topographic and successional gradients (Bratton, 1978). Plots had been placed so
as to not overlap obviously different topographic conditions. Woody plants over 1.0 cm
in diam were recorded by 1.0-cm classes. T h e units of species importance used
throughout the analysis were total m2 of basal area per ha.
Environmental variables. -A soil pit excavated to the top of the C horizon was established on each plot. Depth of organic matter, thickness of A horizon, B horizon, and total unconsolidated soil were noted. Samples were removed from the A and B horizons
for texture and pH analysis, and the percent volume of large stones in the pit was recorded. All soil samples were analyzed for particle size using the Bouyoucos (1963) hydrometer technique and size classes were recorded to the nearest 1% . Water-holding capacity of the A and B horizons was calculated using texture tables by Longwell et. ul.
(1963), A and B horizon thickness, and percent volume stone.
Slope aspect, distance from the plot to surface water, degree of disturbance, nature
of parent material, and percent slope were recorded. Slope aspect was transformed by
the Beers et al. (1966) method. This technique transforms the clockwise increase of azimuth value into a linear sequence from 0.00 (SW aspect) to 2.00 (NE aspect). Topographical type was derived subjectively from maps and in the field (cJ,
Whittaker, 1956)
with types categorized as mesic flat, ravine, cove, protected slope, xeric flat, open slope,
gap and ridge.
An index of site protection was constructed on a scale of 0-100, 0 indicating a plot
with no nearby landforms of greater elevation, and 100 being the plot measured as having the most surrounding land of higher elevation. The variable was labeled "protection"
because it did not measure downslope topography, its meaningfulness being limited to
the surrounding "protecting" topography. The index was developed by calculating the
ratios of the heights of surrounding landforms to the distances of those landforms from
the plot. To selectively sample the protection of a given plot, eight fixed azimuths were
plotted from each plot center at 0, 45, 90, 135, 180, 225, 270 and 315 degrees on topographical maps. The equation
where E R =elevation of the "protecting" landform, E P =elevation of the plot and
D =distance from the plot to the protecting landform, was applied to each plot. Calculations were made for each azimuth and averaged for the plot protection value. In order
to remain consistent with the Beers et al. (1966) transformation's weighing of northern
aspects the 135, 180 and 225 degree azimuths were subjectively weighted by a factor of
1987
CALLAWAY
ET A L . : FOREST
COMMUNITY
ANALYSIS
109
1.5. This weighting resulted in slightly higher values for plots with higher landforms to
the S.
A planimeter and gridform were used on topographical maps to measure the catchment area from which runoff rainfall and snowmelt could drain into the plot. This
value ranged from 1000 m2 (the area of a plot) to an area of greater than 100 ha.
Data analysis. -A sequence of multivariate procedures (after Golden, 1981) was used
to examine the relationships among the forest types. All species were included in all
analyses.
Two-way indicator species analysis. -TWINSPAN, a hierarchical classification -program, was used to initially identify forest types (Hill, 1979b). The function of TWINSPAN is basically to make repeated dichotomies until the sample-units are placed in a
hierarchy ranging from the total collection of sample-units to individual sample-units.
From this hierarchy meaningful units of vegetation, or forest types, must be selected.
Hill (1979b) summarizes the process as (1) the primary ordination (reciprocal averaging) to obtain an initial dichotomy; (2) the refined ordination, derived from indicator
species prominent in the primary ordination, and (3) the indicator ordination. Once a
classification hierarchy was obtained, the forest types were chosen through a series of
comparisons of DECORANA sample-unit groupings, examination of the original data
and consideration of previous work. Ultimately, the decisions were subjective and were
not restricted to a single level of the hierarchy.
&trended correspondence analysis. - T h e precision of the forest type groups obtained from the previous analysis was examined by indirect ordination analysis. The
computer program written by M. 0 . Hill (1979a) was used to execute the procedure.
The major advantages of detrended correspondence analysis (DCA) is the avoidance of
the "arch effect" (Gauch et al., 1977) and the "horseshoe effect" (Kendall, 1971) common
to reciprocal averaging by not permitting systematic relationships between the primary
and secondary axes. Another major fault attributed to reciprocal averaging is the distortion of relative distances between samples on its axes (Hill and Gauch, 1980; Gauch,
1982). DCA summarizes the species composition of samples in various segments of the
gradient and develops an arrangement such that equal differences in species composition correspond to equal differences along the gradient (Hill and Gauch, 1980). The
techniques of replacing orthogonalization with detrending and standardizing the units
of within-samples variance characterize the method of DCA. Environmental causes of
vegetation distribution were investigated by regression and correlation analyses of the
indirect ordinates.
Stepwise discriminant analysis. -Discriminant analysis distinguishes between
groups of observations on the basis of a set of discriminating variables which are expected to differ among the various groups. This is done by forming linear combinations
of the discriminating predictor variables that differ significantly in their group means
(Tatsuoka, 1971; Nie et al., 1975). From previously defined groups (forest types in this
case), sets of discriminant functions are derived stepwise from their linear combinations
of input variables (environmental variables in this case) that present the greatest contrast between the groups (Golden, 1981; Anderson, 1958).
An important use of discriminant analysis is its classification power. After discriminant functions are calculated, they in turn may be used to re-evaluate the groups (forest
types) from which they were derived. This classification process may be used to consider the quality and success of other classification techniques. If a low percentage of
groups has been properly classified, then the original selected variables are poor discriminators by virtue of their commonness among groups (Nie et al., 1975). In this
study discriminant analysis was used to measure the success of the TWINSPAN classification by using environmental variables to establish discriminant functions.
RESULTS
Twelve forest types were chosen from the TWINSPAN hierarchical classification
program by a process including comparisons of DECORANA sample groupings, examination of the original data and consideration of previous work. These forest types
were: white oak-white pine (Quercus alba, Pinus strobus), scarlet oak-yellow pine (Q. coccinea, l? pungens, l? rigida), chestnut oak (Q. prinus), northern red oak-silverbell (Q. rubra,
Halesia carolina), hemlock-silverbell-beech (Tsuga canadensis, H. carolina, Fagus grandifolia),
yellow poplar (or tulip poplar) (Liriodendron tulipifera), hemlock-yellow poplar (Tsuga canadensis, Liriodendron tulipifera), basswood-buckeye (Tilia heterophylla, Aesculus octandra),
northern red oak (Q. rubra), yellow birch-buckeye (Betula alleghaniensis, Aesculus octandra)
and beech (Fagus grandifolia) (Fig. 1, Table 1).
Of the four axes provided by DECORANA two were the most significant and their
ordination scores were tested by linear regression and correlation with all of the environmental variables. The distribution of forest types on the primary axis was best explained by elevation change in the regression analysis, yielding an r2 value of 0.53 (significant at P < 0.001). The primary axis also accounted for more variation in the species
distributions than the other axis, evidenced by an eigenvalue of 0.773 for the primary
axis and an eigenvalue of 0.457 for the secondary axis. The secondary axis was significantly negatively correlated with protection, with r2 = - 0.49 (P < 0.001).
Beech, yellow birch-buckeye, northern red oak, chestnut oak and yellow pine forest
types were clearly separated on the first and second DCA ordinates (Fig. 2). Cove forests overlapped considerably on the ordination. Yellow poplar and hemlock-yellow poplar types occupied similar positions on the first two axes. This overlap is corroborated
by the similarity of direct measurements of elevation, protection and soil textures for
the two types (Table 2).
Environmental variables were used as data for a discriminant analysis on forest type
groups. The resulting functions and their associated coefficients were used to describe
the relationships of the types environmentally and to rank the relative importance of the
rn
I
I
-
White Oak-White Pine
Scarlet Oak-Yellow Pines
1
Chestnut Oak
-
Northern Red Oak-Silverbell
Yellow Poplar
Hemlock-Yellow Poplar
Hemlock-Silverbell-Beech
Northern Red Oak
Basswood - Buckeye
'-r
Buckeye-Yellow Birch
Beech
Fig. 1. -Dendrogram of TWINSPAN hierarchy of forest types from the western GRSM
TABLE
1. -Tree species average basal area (m2/ha) in 12 forest community types. The order of types follows increasing elevation
r,
k3
03
YP
Species
n=
Pinus uirginiana
19 rigida
Quercus coccinea
Q. alba
Q. uelutina
Carya glabra
Oxydendrum arboreum
Nyssa syluatica
Robinia pseudoacacia
Q. prinus
Acer rubrum
Magnolia frasen
Quercus rubra
Liriodendron
tul$ifera
Betula lenta
Fraxinus americana
Halesia carolina
Mapzolia acuminata
Prunus serotina
Tsuga canadensis
Tilia heterophylla
A . saccharum
C. cordiformis
Aesculus octandra
B . alleghaniensis
Fagus grand$olia
12
WOWP
11
SOYP
7
CO
26
NROSil
12
H-SILBeech
9
TP
H-TP
B-B
NRO
YB-B
Bee
11
9
6
11
13
6
U
variables for determining the presence of specific forest types. Scarlet oak-yellow pine,
basswood-buckeye, and beech forest types were all classified correctly by the TWINSPAN program as evaluated by discriminant analysis and the environinental variables
(Table 3). The major classification problems encountered were with the pine-dominated
forest types, all of which were abundant on exceptionally dry sites. TWINSPAN misclassification occurred more frequently for forests which had experienced human disturbance, including hemlock-silverbell-beech and yellow poplar. Discriminant analysis of
the environmental variables correctly classified 74.8% of all of the forest types.
Although five discriminant functions were significant at P < 0.001, the first two accounted for 76.4% of the total variability attributable to the environmental variables
measured. Although the stands were graphed on only the first two discriminant functions, the classification matrix was determined by using all of the calculated functions.
In the first discriminant function elevation was the predominant discriminating variable (Table 4). In contrast to the strong relationship of landform to the second DCA
axis, the second discriminant function was more determined by the pH of the A horizon
and the water-holding capacities of the A and B horizons. Protection is highly correlated with soil pH and water-holding capacity, however, and protection remained prominent in the second function as well as the third, fourth and fifth discriminant functions.
As in the DECORANA graph, the elevation-topographic moisture trend was evident, originating with the xeric yellow pine forest at the low elevation, low p H , exposed
topography corner of the discriminant function graph (Fig. 3). The trend is apparent at
the other extreme in the beech forests at high elevations, and the three cove forest types,
basswood-buckeye, hemlock-yellow poplar, and yellow poplar at the high p H , protected
topography end of the second function. As in the DECORANA graph, the hemlockyellow poplar and the yellow poplar centroids were closely grouped, indicating either a
high degree of habitat overlap or the lack of a discriminating variable for these types.
,;',Northern
',
Red Oak
,
ol',
10
I0
AXIS
\
\
I
Fig. 2. -Distribution of samples on first and second DECORANA ordination axes
TABLE
2. -Environmental characteristics (mean _+ standard error) for forest community types
Yellow
pine
Elevation (m)
Protection
Drainage area (ha)
Total soil depth
(cm)
% Clay in A
%
%
%
%
%
Silt in A
Sand in A
Clay in B
Silt in B
Sand in B
pH of A
pH of B
Water-holding
capacity (cm)
White oakwhite pine
1113 +
0.24 +
1.64 +
79.8 k
44
0.02
0.38
3.0
1201 k 86
0.24 + 0.03
1.54 k 0.31
80.1 k 2.1
27.0 + 1.7
51.5 + 2.6
21.5 + 2.3
32.5 k 1.9
47.7 + 2.9
19.6 + 3.5
5.0 + 0.04
5.2 k 0.04
25.3
51.6
23.3
31.2
48.9
19.3
4.6
4.9
18.4 _+ 1.9
54.5 k 2.2
27.6 + 1.7
22.2 k 2.3
54.5 + 2.8
23.5 + 1.8
5.0 + 0.04
5.1 + 0.04
13.8
49.5
36.1
17.7
51.5
30.6
5.1
5.2
+
k
2.7
2.6
2.1
1.6
3.4
2.4
0.8
0.8
1.7
3.8
2.6
2.4
4.5
3.0
0.05
0.06
19.7
47.2
33.2
21.7
49.3
29.0
4.9
5.2
5.4 k 0.7
4.8
+
1.2
k
2.2
2.6
3.0
2.9
2.8
2.6
0.1
0.1
+
0.9
7.4
k
+
k
+
Hemlocksil-beech
874 + 39
0.21 k 0.02
1.08 _+ 0.02
67.8 + 3.0
27.3
49.2
23.4
30.7
48.8
20.6
5.1
5.2
k
N. red oaksilverbell
73
0.02
0.04
7.1
587 +
0.15 +
0.78 +
66.7 k
+
Chestnut
oak
841 +
0.13 +
0.23 +
61.1 k
572 k 16
0.15 k 0.01
0.75 + 0.28
70.3 k 4.9
k
Scarlet oakyellow pine
16
0.01
0.11
4.4
+
k
k
+
k
k
7.0
+
0.8
k
k
k
k
k
+
k
12.1 k 1.0
9.3
k 1.7
k 1.7
k 2.5
+
1.6
k 1.9
k 2.1
+ 0.07
+ 0.08
+
0.8
TABLE
2. -(Continued)
Yellow
poplar
Elevation (m)
Protection
Drainage Area (ha)
Total soil depth
(cm)
% Clay in A
%
%
%
%
%
Silt in A
Sand in A
Clay in B
Silt in B
Sand in B
Water-holding
capacity (cm)
Hemlockyel. poplar
Basswoodbuckeye
Northern
red oak
Buckeyeyel. birch
Beech
+
1369 k 24
0.09 k 0.02
0.38 k 0.10
73.4 + 4.1
1386 k 26
0.18 + 0.03
1.80 k 0.64
71.0 + 4.1
1575 k 49
0.10
0.01
0.50 + 0.16
50.0 k 5.1
11.7 k 1.5
57.3 + 2.9
30.8 k 2.9
11.8 + 2.1
58.9 + 3.6
29.3 + 3.3
11.0 + 1.7
56.3 + 2.7
32.5 + 1.5
12.3 + 2.1
57.7 k 3.1
30.2 k 1.4
711 +
0.29 +
3.82 +
78.8 k
56
0.03
1.O1
2.3
705 k 54
0.33 k 0.02
5.26 k 1.27
80.1 + 2.4
897
0.38
6.13
80.7
17.3 +
44.3 +
39.3 +
20.1
43.0 +
37.0 k
2.1
2.2
2.6
2.0
2.1
2.8
18.0 + 2.1
42.3 k 1.9
39.9 k 3.4
19.3 k 2.1
43.6
2.5
37.1 + 3.9
15.0 k 1.6
41.4 + 1.2
43.6 k 2.2
16.2
2.0
47.2 + 1.5
36.6 + 2.6
8.7 + 1.0
63.5 k 3.0
27.6 k 3.0
8.1 k 1.6
64.3 + 3.4
27.5 + 3.6
7.9 k 1.2
7.2 k 1.4
11.2 k 2.8
8.3 k 1.0
*
*
+
+
+
24
0.01
1.4
2.8
*
8.2
+
0.8
*
5.7 k 1.3
$
n
,
6
L->
>
z
5
P
$O
5
21
$
TABLE
3. -Matrix of classification success from discriminant analysis of forest types using environmental variables*
Actual
groups
YP
WO-WP
SO-YP
Yellow pine
60.0
20.0
20.0
White oakwhite pine
30.0
50.0
10.0
7.7
Northern red oaksilverbell
Hemlockyellow poplar
B-B
10.0
3.8
80.8
7.7
18.8
72.7
8.5
11.1
66.7
Hemlocksilverbell-beech
Tulip poplar
Predicted group membership
NRO-S
H-S-B
TP
H-TP
100.0
Scarlet oakyellow pine
Chestnut oak
CO
9.1
18.2
22.2
Basswood-buckeye
Northern red oak
Yellow birchbuckeye
Beech
* Percent of grouped forest types correctly classified: 74.8%
11.1
11.1
54.5
11.1
9.1
9.1
66.7
1
m
NRO
YB-B
B
- ..-.
High-elevationforest types. -The spruce-fir association is absent from the western end
of the GRSM where these samples were taken. The three major forest types above 1300
m were northern red oak, buckeye-yellow birch and beech. These stands were environmentally distinct from one another on the basis of elevation and protection (Table 2).
The high-elevation oak stands reported by Whittaker (1956) and absent from Golden's
(1981) study area were represented by the northern red oak type, often composed of
over 90% Quercus rubra canopy coverage (Table 1). The densest of the Q rubra stands
were located along the apex of the state line ridge crest from approximately 1350-1550
m elevation. Quercus rubra was associated in this study with both dry site species and mesic species, both on ridges and in deep coves. This range of distribution has been reported by others (Whittaker, 1956; Golden, 1981; Blackman and Ware, 1982). Blackman and Ware found that the distribution of Q rubra and the more xeric Q. prinus
overlapped extensively on a soil moisture gradient. This complexity was apparent in the
present study. Beech stands, also reported by Whittaker (1956), Golden (1981), Cain
(1937) and Russell (1953), were sampled at higher elevations than any other forest type.
Previous studies have cited ice accumulation, wind funneling, and limited seed dispersal
as causal factors for the presence of these stands (Cain, 1937; Russell, 1953; Whittaker,
1956). It has been proposed that, in the northern parts of its range, Fagus grandifolia is
not usually found on soils where the surface layers are often dry (Harlow and Harrar,
1969). This relationship appears to be true in this study. Although the high-elevation
stands existed on very shallow soils with low water-holding capacities, rainfall during
the growing season was very consistent regardless of precipitation at lower elevations
(Stephens, 1969). Buckeye-yellow birch forests were normally found at the same elevations as northern red oak types but on more protected topography. Duncan's multiple
range analysis of the means of protection and drainage area for the two types showed
them to be significantly different. In general, buckeye-yellow birch stands became freTABLE4. -Standardized coefficients of environmental variables
Discriminant
function 1
Elevation
Protection
Topographical type
Drainage area
Slope aspect
Distance to water
Disturbance class
Parent material
Silt in A horizon
Clay in B horizon
Depth of A horizon
pH of A horizon
Depth of 0 horizon
Total soil depth
Total water-holding capacity
Percentage of variation explained
Discriminant
function 2
quent in coves and ravines from 1200-1300 m , and as the elevation (temperaturemoisture) gradient increased, the forest type extended onto more exposed sites.
Coveforest types.-Species common to the cove forest of the western GRSM are consistent with results reported by Cain (1943), Whittaker (1956) and Golden (1981), although the elevations at which dominant species shift are somewhat different. From approximately 500-800 m, topography with high protection values was dominated by
mixtures of Quercus prinus, Liriodendron tulipifera and Tsuga canadensis. At the top of that
altitudinal range the latter two species dominate with combinations of Tilia heterophylla,
Halesia carolinia and Aesculus octandra. From ca. 800-1000 m, T . heterophylla and A . octandra
predominate with mixtures of ?: canadensis, Frminus americana and Halesia carolina. Above
1000 m Betula alleghaniensis and A . octandra codominate the coves with the abovementioned species, but in steadily decreasing percentages. This was also reported by
Whittaker (1956) and Golden (1981), but the transitions occurred at lower elevations in
the western GRSM. Golden (1981) also reported Tsuga canadensis as a more dominant
cove species in his study area than it was in the western Smokies.
Pine and oak forest types. -Stands dominated by Pinus virginiana were recognized by
Whittaker (1956) but absent from Golden's (1981) data. As was the case for F! rigidadominated types, these stands were found only in disturbed areas with low protection
values. This old-field and xeric correlation was corroborated by Harmon (1980) and
Kuykendall (1978). In similar environmental conditions, but generally on a limestone
substrate., Ouercus alba and f? strobus were dominants as the white oak-white ~ i n eforest
type. Quercus prinus was a common member of the low elevation coves, but as elevation
increased the species spread onto more exposed topography. Above 1000 m Q . prinus
was rarely observed. Pine forests codominated by Q. coccinea were sampled from 5001000 m on the most xeric topography measured. Golden (1974) considered the oak-pine
communities
ecotonal
and samples of the scarlet oak-yellow pine basically correspond
.. . .
.
.
with his evaluation.
-
Basswood- Buckeye
Yellow Poplar
Hemlock-Yellow Poplar
Northern Red Oak-Silverbell
Chestnut Oak
Buckeye -Yellow Birch
Hemlock - Silverbell- Beech
-2
@Yellow Pines
Northern Red Oak
-
Scarlet OakYellow Pines
Beech
White OokWhite Pine
-1
-4
-2
o
DF-ONE
2
4
Fig. 3.-Centroids of forest community types along the first two discriminant functions
(DF) based on environmental variables
As in preceding studies, forest types overlapped considerably in species composition
and habitat (Whittaker, 1956; Golden, 1974, 1981). This continuity of forest types was
in part due to the continuity of the predominant environmental variable, elevation. Elevation, in turn, is actually an environmental complex including temperature and moisture. This complexity is apparent in community relationships throughout the Smoky
Mountains. For example, the three high-elevation forest types are composed primarily
of species found at lower elevations only in protected cove and ravine topography. Aesculus octandra, Qwrcus rubra, Betula alleghaniensis, Acer saccharum and Fagus grandifolia dominated at the higher elevation under different topographic conditions, but as elevation
decreased these species almost exclusively occupied sheltered topography. Although all
of the above species followed this basic pattern, the communities in which they were
found changed with elevation. In other words, entire communities did not move with
changing elevation and topography. For example, Aesculus octandra was associated with
B . alleghaniensis in highly exposed ridgetop plots but at lower elevations its primary associate was Tilia heterophylla in deep concave topography. Quercm rubra was another prime
example of this community sharing. Querctu rubra was common on the most exposed
ridges measured at high elevations (in almost pure stands), in protected cove positions
with Liriodendron tulipifera, Halesia carolina and Tilia heterophylla, and with Q. prinus in dry
southerly positions. This conflicts with the traditional assumption that Q. rubra is an essentially mesic or submesic species (Whittaker, 1956; Mowbray, 1966; Golden, 1974).
This finding is substantiated by a study of Q. rubra and its association with Q. prinus
(Blackman and Ware, 1982). Stands of Q, rubra could not be correlated with moisture
conditions, and samples occurred at both extremes of a soil moisture gradient.
Although the continuity of elevation and topography resulted in highly overlapping
community distributions, some forest types were relatively distinct on the DECORANA axes (Fig. 1). Beech, northern red oak-silverbell, northern red oak, buckeyeyellow birch, and the yellow pines forest types were clearly separated. It is questionable,
however, whether this distinctness is a result of environmental control or is an artifact of
the location of plots so that the overlap is not sampled.
Changes of land shape have been considered by previous researchers as a dominant
condition influencing vegetation distribution in the Smokies, but they were not quantified. Thus the importance of land shape was measured in fragmented pieces, such as
distance from ridgetop, slope length, distance from valley bottom and other similar
measurements. Protection, although not measuring all aspects of topography, proved to
be a significant step towards quantifying land shapes, and thus toward understanding
vegetation distribution. As for elevation, protection is most likely a complex variable including relationships to isolation, drainage, soil particle movement, temperature and
wind. We assumed that the most influential aspects of topography are products of the
landforms at higher elevations than the plot. Although this assumption may be true, it
is undoubtedly incomplete. Slope shape at elevations lower than the plot is partially responsible for air flow, water drainage and soil particle movement. Despite the problems
inherent in the variable, protection, it has proven useful in this study, and it has proven
highly significant for prediction of forest productivity (Callaway, 1983).
The complexity of the variable, protection, was apparent in the results of the discriminant analysis. Although protection was an important variable in four of the five
functions, it was replaced by the p H of the A horizon and the total water-holding capacity of the soil as the most important variable in the second discriminant function. Both
of these variables are highly correlated with protection. It is likely that topographic
shape is partially responsible for the soil conditions at a given site. Thus it is possible
that elements of the environmental complex inherent in landform are being measured.
Golden (1981) measured variation in soil p H with topographic change and, as in this
study, found the p H of the A horizon was a major variable correlated with vegetation
distribution.
Human disturbance appeared to be an important factor in the distribution of
Smoky Mountains vegetation but was very difficult to quantify. All of the yellow pine
and white oak-white pine plots had been disturbed by farming or logging, as were high
percentages of the yellow poplar and hemlock-silverbell-beech forests. In the case of the
first two there are no undisturbed areas of similar habitat with which to compare them.
Because of the lack of comparative vegetation with no or little human disturbance, this
study can determine little about the influence of disturbance on the yellow pine and
white oak-white pine forest types. Alternatively, the abundance of undisturbed sites similar to those of yellow poplar and hemlock-silverbell-beech indicate that other communities may be more common without the influence of human disturbance. Of the six yellow poplar plots that overlapped most with hemlock-yellow poplar forest types (numbers
7 and 8 in Fig. 2) five were disturbed. This indicates that human disturbance is influential in the development of yellow poplar forest types from what potentially would be
hemlock-yellow poplar. A similar situation can be observed with the hemlock-silverbellbeech forest type. Almost 80% of these plots were disturbed and the forest type overlaps
highly with both the basswood-buckeye forest type and the northern red oak-silverbell
forest type (Fig. 2). This indicates that the hemlock-silverbell-beech forest type may be a
product of human disturbance and without disturbance the community composition
would be different. Conversely, human disturbance may convert sites supporting
hemlock-yellow poplar to yellow poplar forest types, and sites supporting basswoodbuckeye and northern red oak-silverbell communities into different groups of species
better classified as hemlock-silverbell-beech communities.
Human disturbance has been cited as influential to other forest types in the Smokies. Harmon (1980) found that yellow pines increased their coverage in fire-disturbed
sites. Forest composition in the GRSM has also been changed by the widespread loss of
the dominant Cmtanea dentata. Former chestnut-dominated stands now show increased
cover of Querctu printu, Q. rubra and Acer rubrum (Woods. and Shanks, 1959). Golden
(1981) found that the inclusion of dead Cmtanea dentata remains was the most significant
environmental variable entered in his stepwise multiple discriminant analysis. He attributed the significance to the opening of the canopy, which encouraged the growth of
specific shade-intolerant species, in turn establishing specific forest types. In previous
studies, as well as in this one, disturbance worked synergistically with topography, soil
characteristics and elevation to determine vegetation distribution in the Smoky Mountains. Because there has been some disturbance in all communities of the GRSM, vegetation results show some degree of increased variation.
Acknowledgments.-This study was supported by the U.S. Forest Service, Southeastern Forest
Experiment Station, under Cooperative Agreement USDA 18-884 with the University of Tennessee, Knoxville, as a part of a program to predict forest pattern and growth from variables
recognized by remote sensing techniques. Much of the vegetative and environmental data was
provided by the Uplands Field Research Laboratory of the National Park Service.
LITERATURE
CITED
ANDERSON,
T. W. 1958. An introduction to multivariate statistical analysis. John Wiley and
Sons, Inc., New York. 374 p.
ARENDS,
E. Vegetation patterns a half century following the chestnut blight in the Great Smoky
Mountains National Park. M.S. Thesis, University of Tennessee, Knoxville. 79 p.
BEERS,
T. W., P. E. DRESS
A N D L. C. WENSEL.
1966. Aspect transformation in site productivity
research. J. For, 64:691-692.
BLACKMAN,
D. A N D S. WARE.1982. Soil moisture and the distribution of Quercus prinus and
Quercus rubra. Cmtanea, 47:360-367.
BRATTON.
S. F. 1978. Vegetation survey Great Smoky Mountains National Park. National Park
Service, Southeast Regional Office, Uplands Field Research Laboratory. 53 p.
Bou~oucos,G. J. 1963. Directions for making mechanical analysis of soil by the hydrometer
method. Soil Sci., 42:225-229.
CAIN,S. A. 1930. Certain floristic affinities of the trees and shrubs of the Great Smoky Mountains and vicinity. Butler Uniu. Bot. Stud. , 1 : 129-150.
.
1937. A preliminary guide to the Greenbrier-Brushy Mountain nature trail, The
Great Smoky Mountains National Park. University of Tennessee, Knoxville. 29 p.
.
1943. The Tertiary character of the cove hardwood forests of the Great Smoky Mountains National Park. Bull. Torrey Bot. Club, 70:213-235.
CALLAWAY,
R . M . 1983. Prediction of growth in classified forest stands in the western Great
Smoky Mountains National Park. M.S. Thesis, University of Tennessee, Knoxville.
110 p.
GAUCH,
H . G., JR. 1982. Multivariate analysis in community ecology. Cambridge Univ. Press,
New York. 289 p.
,
R. H . Whittaker and VT. R. Wentworth. 1977. A comparative study of reciprocal averaging and other ordination techniques. J. Ecol., 65: 157-174.
GOLDEN,
M . S. 1974. Forest vegetation and site relationships in the central portion of the
Great Smoky Mountains National Park. Ph.D. Dissertation, University of Tennessee,
Knoxville. 275 p.
, 1981. An integrated multivariate analysis of forest communities in the central Great
Smoky Mountains. Am. Midl. Nut., 106:37-53.
HARLOW,
W. M . AND E. S. HARRAR.
1969. Textbook of dendrology. McGraw-Hill, New York.
512 p.
HARMON,
E. H. 1980. The influence of fire and site factors on vegetation pattern and process:
a case study of the western portion of the Great Smoky Mountains National Park.
M .S. Thesis, University of Tennessee, Knoxville. 170 p.
HILL,M . 0 . 1979a. DECORANA, a fortran program for detrended correspondence analysis
and reciprocal averaging. Cornell Univ., Ithaca, New York. 52 p.
.
1979b. TWINSPAN, a fortran program for arranging multivariate data in an ordered two-way table for classification of individuals and attributes. Cornell Univ.,
Ithaca, New York. 90 p.
A
N
D H . G. GAUCH,
JR. 1980. Detrended correspondence analysis, an improved ordination technique. Egetatio, 42:47-58.
D. G. 1971. Seration from abundance matrices. Mathematics in the archaeological
KENDALL,
and historical sciences. Edinburgh Univ., Edinburgh. 252 p. KUYKENDALL,
N. W. 1978. Composition and structure of replacement forest stands following
southern pine beetle infestations as related to selected variables in the Great Smoky
Mountains. M.S. Thesis, University of Tennessee, Knoxville. 122 p.
A ND M . E. SPRINGER.
1963. Moisture characteristics of TennesLONGWELL,
T. J., W. I. PARKS
see soils. Enn. Agric. Exp. Stn. Bull. 367. 46 p.
T. B. 1966. Vegetation gradients in the Bearwallow Gorge of the Blue Ridge escarpMOWBRAY,
ment. J. Elirha Mitchell Sci. Soc., 82: 138-149.
NIE, N. H . , C . H. HULL,J. H . JENKINS,
K. STEINBRENNER
A N D D. H . BENT.1975. Statistical
package for the social sciences, ed. 2. McGraw-Hill, New York. 675 p.
PEET,R . K. 1980. Ordination as a tool for analyzing complex data sets. Egetatio, 42:171-174.
N. H . 1953. The beech gaps of the Great Smoky Mountains. E c o l o ~ 3, 4:366-374. RUSSELL,
L. A. 1969. A comparison of climatic elements at four elevations in the Great
STEPHENS,
Smoky Mountains. M.S. Thesis, University of Tennessee, Knoxville. 119 p. TATSUOKA,
M . M . 1971. Multivariate analysis. John Wiley and Sons, Inc., New York. 310 p.
R . H. 1956. Vegetation of the Great Smoky Mountains. Ecol. Monogr., 26:l-80.
WHITTAKER,
WOODS,F. W. A N D R . E. SHANKS.
1959. Natural replacement of chestnut by other species in
the Great Smoky Mountains National Park. Ecology, 40:349-361. YONG,R . W. A N D B. P. WARKENIN.
1966.
Introduction to soil behavior. MacMillan, New York.
451 p.